Voluntary Disclosure of Evaded Taxes Increasing Revenues, or Increasing Incentives to Evade?

Transcription

1 Voluntary Disclosure of Evaded Taxes Increasing Revenues, or Increasing Incentives to Evade? Dominika Langenmayr University of Munich February 16, 2015 Abstract Many countries apply lower fines to tax evading individuals when they voluntarily disclose the tax evasion they committed. I model such voluntary disclosure mechanisms theoretically and show that while such mechanisms increase the incentive to evade taxes, they nevertheless increase tax revenues net of administrative costs. I then test the effects of voluntary disclosure in two separate empirical analyses. First, I confirm that voluntary disclosure mechanisms increase tax evasion, using the introduction of the 2009 offshore voluntary disclosure program in the U.S. for identification. Second, I quantify the tax revenues of voluntary disclosures by considering how some state-level governments in Germany bought whistle-blower data from foreign bank employees, thereby increasing the detection probability and the usage of voluntary disclosures. Keywords: Tax evasion, voluntary disclosure, self-reporting JEL Classification: H26, K42, H24 I am grateful to Joel Slemrod (discussant), Stephanie Sikes (discussant), Nadja Dwenger, Andreas Haufler, Gareth Myles and participants of the 2014 NBER Trans-Atlantic Public Economic Seminar in Vienna and the public economics seminar of the University of Munich for very helpful comments. My thanks also go to Lisa Eßbaumer for excellent research assistance, and to the Economics Department of UC San Diego for their hospitality. Financial support from the German Research Foundation (LA 3565/1-1) is gratefully acknowledged. dominika.langenmayr@econ.lmu.de.

2 1 Introduction Households worldwide hold about 8% of their total financial wealth, almost U.S.-$ 6 trillion, in tax havens (Zucman, 2013). Correspondingly, tax authorities forego high tax revenues: The United States forego tax revenues of around $ 70 billion annually due to personal income tax evasion via offshore accounts (Gravelle, 2009). The need for tax revenues in the wake of the financial crisis has now rekindled governments efforts to curb such income tax evasion. Principally, governments can fight tax evasion by individuals who hold their wealth offshore in two ways. First, they can negotiate with tax havens to share information regarding foreigners accounts. An example is the recent agreement between the United States and Switzerland forcing Swiss banks to provide information on accounts owned by U.S. citizens. However, such treaties are not very effective, as tax evaders rather shift their funds to different tax havens instead of repatriating them (Johannesen and Zucman, 2014). Second, governments can set incentives for individual taxpayers to declare foreign wealth and the tax evaded on it. Many countries incentivize individuals to come clean by a voluntary disclosure mechanism. Usually, the prerequisite is to report all foreign asset holdings. The income on these assets is then taxed retroactively at the standard tax rate, but no or a reduced fine is levied. Only individuals not yet under investigation for tax evasion can profit from such programs. Voluntary disclosure programs exist in many countries (see Table 1 for an overview), and are often part of the general law and for an unlimited period. However, some commentators fear that the option of voluntary disclosure increases the incidence of tax evasion, as these programs offer the possibility to escape high punishments if individuals decide that the probability of detection has increased. [Table 1 about here.] The economic literature has so far barely studied voluntary disclosure programs. Using both a theoretical model and empirical tests, the paper aims to shed some light on this topic. First, I ask how the existence of a voluntary disclosure program affects individuals tax evasion decision. In both the theoretical model and the empirical test I show that the existence of such a program increases tax evasion. Second, I consider the government s point of view, studying whether the tax authorities should offer voluntary disclosure, despite the increase in tax evasion it causes. In my model, governments 2

3 should offer voluntary disclosure only when a disclosure lowers the administrative costs related to assessing taxes of evaders. I then confirm the importance of administrative costs in a survey of German competent local tax authorities. In a second empirical test I quantify the revenues brought in by voluntary disclosure in Germany. Lastly, I analyze how governments should fine tax evaders after a voluntary disclosure. In more detail, my theoretical model frames tax evasion as a rational choice of individuals that bear a moral (psychic) cost when evading taxes. There is ex-ante uncertainty about the probability of being caught and fined, and individuals have the possibility to voluntarily disclose the tax evasion they committed after the detection probability is revealed. In equilibrium, the individuals with the lowest moral cost will evade taxes, those with intermediate moral costs will first evade taxes but voluntarily disclose later when the detection probability is high, and those with the highest moral costs will never evade taxes. In this model I show that the existence of voluntary disclosure increases the number of individuals who evade taxes. This result arises as voluntary disclosure allows individuals to better differentiate their actions according to the detection probability. I later test this result empirically, using the introduction of the first Offshore Voluntary Disclosure Program in the U.S. in 2009 for identification. Employing a synthetic control approach, I analyze how U.S. deposits in offshore banking centers have changed compared to deposits from other countries. This analysis confirms that the existence of a voluntary disclosure program indeed increases tax evasion, in line with the theoretical model. I also model how the government should employ voluntary disclosure. In the model, voluntary disclosure increases net tax revenue if there are administrative costs of fining tax evaders in the absence of a voluntary disclosure. I confirm the importance of these administrative cost savings in a survey among German competent local tax authorities. In the model, government optimally set voluntary disclosure fines by trading off higher tax evasion with these savings in administrative costs. In a last step, I study shocks to detection probabilities in Germany to gauge the additional revenue of a voluntary disclosure. My estimates suggest that one voluntary disclosure brings in around e26,000 on average. Several strands of literature are relevant to this paper. First, there is a large literature on tax evasion by individuals (see Slemrod (2007) for an overview). The theory goes back to Allingham and Sandmo (1972) and Yitzhaki (1974), who model tax eva- 3

4 sion analogous to portfolio choice. Sandmo (2005) provides a review of this line of literature. Despite the obvious difficulties to measure tax evasion, there is also a large empirical literature, which Alm (2012) summarizes. To my knowledge, no paper studies a voluntary disclosure program as described above. However, there is some literature on tax amnesties, which are short-run programs (often about three months long) that usually do not fine tax evaders. Also in contrast to voluntary disclosures, tax amnesties often include those already under investigation for tax evasion and allow only a partial reporting of prior tax evasion. In this literature, Malik and Schwab (1991) propose a model with uncertainty about the disutility from tax evasion to explain why individuals take up the offer of a tax amnesty (which they never would in the standard Allingham-Sandmo model). Alm and Beck (1990) set up a prospect theory model in which the share of evaded tax that is declared in the amnesty is the main decision parameter. Stella (1991) discusses the interaction between future enforcement and tax amnesties, predicting that amnesties are unlikely to generate additional revenue. Alm and Beck (1993) confirm this result empirically in a time-series analysis. 1 Closer to this paper is an analysis by Andreoni (1991), who asks how a permanent tax amnesty (in effect, a voluntary disclosure program in the sense discussed above) would affect the efficiency and equity of the tax system. He proposes a model in which people use the amnesty when shocks to their consumption make them unwilling to bear the risk of audit. In this model, the tax amnesty acts similar to social insurance, allowing those in bad luck to eliminate some of their risk. He does not consider administrative costs or the optimal fine set by the government, but assumes (as common in this 1 Other papers have studied behavioural responses of individuals to temporary decreases of other taxes. For example, Agarwal et al. (2013) show very large behavioral responses to sales tax holidays, brief periods in which sales taxes are reduced or eliminated. They find little evidence of substitution across products or over time. In contrast, Cole (2009) estimates that sales tax revenues decrease by up to 8% during tax holiday months. Studying a stamp duty holiday in the UK, Besley et al. (2014) similarly find that the tax holiday mostly leads to short-term retiming of transactions. However, due to the specific situation of prior tax evasion, it is unclear how well these results transfer to voluntary disclosure. 4

5 literature) that there is no fine after a disclosure. 2 In all, voluntary disclosure programs with their current characteristics (e.g. requiring full disclosure, and having a specifically chosen fine as their most prominent characteristic) have so far been an understudied aspect of tax evasion. This paper aims to shed some light on them. Section 2 provides the theoretical model. Section 3 empirically tests some aspects of voluntary disclosure. Section 4 concludes. 2 Model 2.1 Framework To illustrate the consequences of voluntary disclosure, I set up a model in which individuals may evade capital income taxes by transferring money to an offshore account. The government can set incentives for tax evaders to come clean by offering them the possibility to voluntarily disclose the tax evasion they committed. Offshore income indicated in a voluntary disclosure is then taxed, and fined at a rate chosen by the government. From the government s point of view, a voluntary disclosure has two main advantages. First, it detects tax evasion that potentially would not have been exposed otherwise. Second, a voluntary disclosure saves the government administrative costs, such as the cost of time spent collecting information from less-than-cooperative offshore banks. 3 Individuals in the model face ex-ante uncertainty about detection probabilities. This uncertainty reflects, for example, that there is a certain probability that an informant offers the government information about offshore accounts. Figure 1 clarifies the real-world significance of changes in detection probabilities using the example of 2 Some further papers study the optimal self-reporting of violations of the law in a non-tax context. A first contribution is Kaplow and Shavell (1994), who show that self-reporting increases welfare as it saves enforcement resources and reduces uncertainty for individuals facing potential sanctions. Their model has been extended to consider ex-post asymmetric information (Feess and Heesen, 2002) or self-reporting at different stages of an investigation (Feess and Walzl, 2005). 3 Tax advisors and tax authorities have confirmed in private discussions that a voluntary disclosure often has several hundred to thousand pages, as the individual not only has to disclose all capital income over a certain period, but also prove that the disclosure is complete. 5

6 Figure 1: Voluntary Disclosures per Quarter in Germany Voluntary Disclosures per quarter in Germany, Q Q Graph based on information from state finance ministries, data for is extrapolated based on information for Lower Saxony and Schleswig-Holstein. Germany, which has bought at a large scale whistle-blower information offered by former bank employees in tax havens. The acquisition of such data from February 2010 onward was widely discussed in the media. In 2011, it emerged that Germany and Switzerland had negotiated a tax treaty under which undeclared accounts of German nationals in Switzerland would be subject to a one-time tax payment. This single tax payment was supposed to be collected anonymously and to exempt the account holder from prosecution for tax evasion committed in the past. However, in November 2012, the upper house of the German parliament did not pass this tax treaty, thus making voluntary disclosure again the only possibility to come clean on past tax evasion. 4 The model reflects such changes in the underlying detection probabilities. With probability q, a high detection probability p H occurs (e.g. because the government receives whistle-blower information). Correspondingly, with probability 1 q there is no leak and the detection probability is low (p L ). This uncertainty not only reflects 4 See Pfisterer (2013) for an overview of these developments. 6

7 the real-world facts described above, but is also necessary for the model, as rational individuals will only voluntarily disclose tax evasion they optimally chose to commit earlier if they have received new information. 5 Not all individuals have the same willingness to evade taxes. Kleven et al. (2011) show that among Danish taxpayers who self-report their income (and thus have the opportunity to evade taxes), less than 40% actually evade taxes. I model such heterogeneity among individuals with a moral cost of tax evasion (α i [0; A]), which is specific to the individual. In equilibrium, there will be three different groups of individuals: First, a group of non-evaders, who have high moral costs and never evade taxes; second, disclosers, who evade taxes but voluntarily disclose if the detection probability is high; and lastly evaders, who evade no matter which detection probability occurs. Individuals decide about tax evasion and voluntary disclosure by maximizing their expected utility. 6 I assume risk neutral individuals, whose utility is U i = y τ is 1α i. (1) y is the pre-tax capital income, τ is is the tax (and fine) payment that depends on individual i s tax evasion and disclosure decisions as well as the state of the world s, and 1 is an indicator function that is equal to one if the individual evades taxes and zero otherwise. Due to the linear structure of the utility function, it is never optimal to declare a share of the true income. In the main model, I assume risk neutral, heterogeneous individuals as these assumptions allow a tractable model that fits well to the stylized facts described above. To provide robustness and make sure that risk averse individuals would not change 5 I treat the two potential detection probability as exogenous, as the government can affect them only in the medium to long term. Tax authorities usually do not hire short-term workers as employees have to handle sensitive tax data, and training new employees takes time. Moreover, strong job protection rules in the public sector make short-term adjustments very costly. Other short term actions, such as negotiating new information sharing agreements with tax havens, can be understood as an occurrence of p H. 6 There is also a part of the literature that does not rely on expected utility theory. Alm et al. (1992) show in an experiment that some individuals overweight the low probability of audit. Dhami and Al-Nowaihi (2010) model such behavior using prospect theory and predict a positive relationship between tax rates and tax evasion. However, when testing whether expected utility theory or prospect theory provide a better explanation of individuals behavior regarding tax evasion, King and Sheffrin (2002) find experimental evidence in favor of expected utility theory. 7

8 the main implications of the model, Appendix 1 provides a model that focusses on the tax evasion decision of risk averse individuals in the presence of a voluntary disclosure mechanism. In that model, every individual evades some (but not all) tax. In the presence of voluntary disclosure, all individuals evade more taxes than in the benchmark case without voluntary disclosure. In sum, the main results concerning aggregate tax evasion and tax revenues are very similar to the main model. Such a model with risk aversion could be extended to take heterogeneity in the degree of risk aversion into account (as discussed in Appendix 1). In such a model, the most risk-averse individuals would never evade taxes; individuals with an intermediate degree of risk-aversion would evade taxes but disclose when the detection probability is high; and the least risk-averse individuals would always evade taxes, and never voluntarily disclose. The introduction of a voluntary disclosure program then affects tax evasion behavior at two margins: First, it increases the amount of tax each tax evader evades, and it increases the number of individuals who evade taxes. Let me now return to the model with risk-neutral individuals and describe the tax system. Individuals are liable to pay taxes at the statutory rate t. They can evade this tax by hiding their money in an offshore account and not declaring the income derived from it. If the tax authorities detect the tax evasion, the individual pays a fine F > 1 that is proportional to the evaded tax. I treat this fine for tax evasion, F, as exogenous, assuming that it is set in an appropriate relation to punishments for other crimes. 7 The government may allow voluntary disclosures of prior tax evasion. As is common practice, a voluntary disclosure requires that the tax evader reports all income on which he or she evaded taxes. It is costly for individuals to prepare a voluntary disclosure, as they have to collect all information necessary to assess their taxes. I model this with a compliance cost c c that arises when preparing a voluntary disclosure. The government retroactively collects taxes on the income declared in the voluntary disclosure. Additionally, it imposes a fine f (1 f F ), which is proportional to the evaded tax. The government sets this fine to maximize revenues. I assume that a voluntary disclosure clears the conscience of the individual, i.e. that after a voluntary disclosure the individual no longer has moral qualms about the tax evasion committed earlier. 7 Since the analysis of Kolm (1973), it is well known that with positive marginal costs of auditing, the government optimally sets the fine for tax evasion to the maximum level that is in line with moral and legal constraints. This is the implicit assumption in the model presented here. 8

9 Figure 2: Game tree 1 st stage: Government sets f 2 nd stage: Individuals... evade don t evade 3 rd stage: Nature q 1 q q 1 q Individuals 4 th stage: p L disclose don t discl. p H disclose don t discl. p L disclose don t discl. p H disclose don t discl. 5 th stage: Audits are carried out, taxes and fines are paid. The government incurs administrative costs c a for each tax evading individual, as it checks and audits the tax return and collects information from offshore banks. For the tax authorities, these costs are significantly higher than the costs that a tax evader incurs when preparing a voluntary disclosure: The tax authorities have to investigate to detect all foreign accounts and asset holdings, and then have to obtain detailed information on the movements of funds from (perhaps less-than-cooperative) offshore banks. In contrast, the individuals themselves either have all this information already available, or can (as the account holders) easily request it from their banks. Voluntary disclosure will only take place in equilibrium if there are rents to share, i.e. when the costs of preparing a voluntary disclosure (c c ) are sufficiently lower than the administrative costs of fining tax evaders (c a ). On the other side, administrative costs may not be so high that fining tax evaders is no longer worthwhile for the government. To make sure that these conditions are met, I assume that c c (1 q)(p h p L ) ca < F ty. (2) This condition assures that individuals are willing to disclose (left part), and that the government is willing to fine evaders despite the administrative costs (right part). Figure 2 describes the stages of the game in detail. First, the government sets the 9

10 voluntary disclosure fine f. In the second stage, individuals decide whether to evade taxes. They anticipate that nature will draw the detection probability in the next stage. After the detection probability is revealed, individuals may have the option to voluntarily disclose the tax evasion they committed. Lastly, the government audits some taxpayers, and individuals accordingly pay taxes and fines. 2.2 Benchmark Without Voluntary Disclosure As a benchmark, consider first the case when voluntary disclosure is not possible, i.e. the game without the fourth stage. Individuals then base their evasion decision on the expected detection probability, p, with p = qp H + (1 q)p L. Their expected payoff depends on whether they evade taxes or not: EU 0 (Don t evade) = y ty EU 0 (Evade) = y pf ty α i Comparing the expected utility in these two cases shows that individuals with a moral cost α i < α 0 evade taxes, with α 0 given by α 0 = ty (1 pf ). (3) The number of evaders is higher the higher the potential gain from tax evasion (ty), and lower the higher the expected fine ( pf ty). 8 For later use, tax revenues net of administrative costs, T 0, are α 0 T 0 = p (F ty c a ) dg(α i ) + 0 A α 0 tydg(α i ). (4) The first term denotes the taxes and fines net of administrative costs that the tax authorities collect from evaders; the second term are taxes that non-evaders pay. 2.3 Voluntary Disclosure A voluntary disclosure implies that an individual reports all income on which he or she evaded taxes to the authorities. In most countries voluntary disclosures are 8 As common in the literature I assume that pf < 1, i.e. that tax evasion is worthwhile in expectation. 10

11 associated with a fine (see Table 1 for details). The voluntary disclosure fine f 1 is lower than the fine for tax evasion (f F ), and in some countries no fine is levied (f = 1). The individuals choice set is (Evade, Disclose), (Evade, Don t disclose), (Don t evade, Disclose), (Don t evade, Don t disclose), and the corresponding payoffs are EU(Evade, Disclose) = y fty c c EU(Evade, Don t disclose) = y p j F ty α i EU(Don t evade, Disclose) = y ty c c EU(Don t evade, Don t disclose) = y ty, where p j is the detection probability drawn by nature in the third stage. Knowing these different outcomes, individuals decide about tax evasion anticipating the full sequence of events. The government, in turn, takes individuals decisions into account and sets the voluntary disclosure fine accordingly. I solve the game by backward induction. 4th stage: Disclosure decision. Consider first the disclosure decision of an individual who has not evaded taxes in the second stage. As a voluntary disclosure does not change this individual s tax payment, but entails positive compliance costs c c, it is never optimal. Thus, (Don t evade, Disclose) is a dominated strategy. Voluntary disclosure is only attractive for individuals who have evaded taxes, and who may use it to lower their expected fine or avoid the moral cost associated with tax evasion. Comparing EU(Evade, Disclose) and EU(Evade, Don t disclose) shows that for a detection probability p j, j [L, H], voluntary disclosure is optimal for tax evaders with moral costs α i α vd j, with α vd j = ty(f p j F ) + c c. (5) This cutoff is lower (that is, more individuals voluntarily disclose) when nature draws the high detection probability ( αh vd < ) αvd L. Moreover, more individuals voluntarily disclose when the fine associated with voluntary disclosure is low. But even when there is no fine after a voluntary disclosure (f = 1), not all individuals will voluntarily disclose even if the detection probability is high, as long as there still is an expected gain from tax evasion. Higher compliance costs associated with preparing the voluntary disclosure also make it less attractive. 11

12 Lemma 1 It is never optimal for individuals who have not evaded taxes to voluntarily disclose. Tax evaders disclose when the high (low) detection probability occurs if they have moral costs of tax evasion above α vd H (αvd L ) as given by Eq. (5). 3rd stage: Detection probability. Nature draws the high detection probability p H with probability q, and the low detection probability p L with probability 1 q. 2nd stage: Evasion decision. The evasion decision depends on the individual s moral cost of tax evasion. First, consider individuals with moral costs α i [0, α vd H ), who will not voluntarily disclose in stage four. Comparing their expected utilities EU(Evade, Don t disclose) and EU(Don t evade, Don t disclose) shows that it is optimal for these individuals to evade taxes. Thus, their equilibrium strategy is {Evade, (Don t disclose if p L ), (Don t disclose if p H )}. Off the equilibrium path, i.e. if they had chosen not to evade taxes, they would also choose {(Don t disclose if p L ), (Don t disclose if p H )}. Next, consider individuals with moral costs α i [α vd H, αvd L ). We know that these individuals choose {(Don t disclose if p L ), (Disclose if p H )} if they evaded in the first stage, and {(Don t disclose if p L ), (Don t disclose if p H )} if they did not evade in the first stage. Comparing the expected utilities for these two options shows that tax evasion is only optimal for individuals with moral costs α i < α t, with α t = ty 1 qf (1 q)p LF 1 q q 1 q cc. (6) Note that α t is indeed between αh vd and αvd L : αt αl vd holds as cc (f 1)ty. 9 α t αh vd when ty [1 f + (1 q)(p H p L )F ] c c. After deriving the equilibrium fine it will become clear that this condition holds when the compliance costs of the individual preparing the voluntary disclosure are sufficiently lower than the administrative costs of the tax authority, i.e. when (2) is met. Lastly, we know that individuals with α i [αl vd, A] always disclose after evading taxes, no matter which detection probability arises. They do so because of their high moral cost of tax evasion, and because of this they also prefer not to evade taxes in the 2nd stage. Lemma 2 summarizes the tax evasion decision for all groups of individuals. 9 To see this, note that f > 1 and c c > 0. 12

13 Lemma 2 Individuals tax evasion decision depends on their moral cost of tax evasion. Individuals with α i < α t choose to evade taxes, with α t given by (6). Combining these insights allows to describe the full strategies of each group of individuals: Lemma 3 Individuals optimal strategy depends on their moral cost of tax evasion. Individuals with α i [ 0, α vd H ) choose {Evade, (Don t disclose if pl ), (Don t disclose if p H )}. Off equilibrium, i.e. if they had chosen not to evade taxes in the second stage, their disclosure decision would also be {(Don t disclose if p L ), (Don t disclose if p H )}. α i [ α vd H, αt) choose {Evade, (Don t disclose if p L ), (Disclose if p H )}. Off equilibrium, their disclosure decision would be {(Don t disclose if p L ), (Don t disclose if p H )}. α i [ ) α t, αl vd choose {Don t evade, (Don t disclose if pl ), (Don t disclose if p H )}. Off equilibrium, i.e. if they had chosen to evade taxes in the second stage, their disclosure decision would be {(Don t disclose if p L ), (Disclose if p H )}. α i [ α vd L, A] also choose {Don t evade, (Don t disclose if p L ), (Don t disclose if p H )}. However, off equilibrium, their disclosure decision would be {(Disclose if p L ), (Disclose if p H )}. Figure 3 summarizes these equilibrium strategies for both a world with and without voluntary disclosure. It shows that when nature draws the low detection probability, it is never rational to voluntarily disclose in equilibrium after all, the same individual chose to evade taxes when it was still unclear whether the low or the high detection probability would occur. However, if nature draws the high detection probability, evaders with relatively high moral costs of tax evasion may opt for a voluntary disclosure, preferring a certain, but lower, fine payment and clear conscience over the tax saving with the risk of a high fine if evasion is detected. 1st stage: Government. In the first stage, the government sets the voluntary disclosure fine to maximize expected tax revenue net of administrative costs. Ex ante, i.e. 13

14 Figure 3: Equilibrium behavior of individuals with different moral costs α i. Without voluntary disclosure program: 0 α 0 Evade Don t evade α i With voluntary disclosure program: 0 αh vd α t αl vd Evade Don t discl. if p L Don t discl. if p H Evade Don t discl. if p L Discl. if p H Don t evade Don t discl. if p L Don t discl. if p H Don t evade Don t discl. if p L Don t discl. if p H α i before the detection probability is revealed, expected tax revenues net of administrative costs are T = α vd H 0 ty dg(α i ). α t A p (F ty c a ) dg(α i )+ [qfty + (1 q)p L (F ty c a )] dg(α i )+ α vd H α t (7) The first term refers to the revenue collected from evaders, the second term to the expected revenue from those who voluntarily disclose when the detection probability is high, and the last term to the revenue collected from non-evaders. Note that no administrative costs arise after a voluntary disclosure, as the disclosure has to contain all information necessary for assessing the tax liability. 10 The government sets the fine that applies after a voluntary disclosure to maximize its expected tax revenue. Assuming that there is a mass M of taxpayers with moral costs α i distributed uniformly in the interval [0, A], Eq. (7) can be rewritten as T = p (F ty c a ) αvd H M A + [qfty + (1 q)p L(F ty c a )] (αt α vd H )M A + ty (A αt )M, A (8) with α vd H and αt given by Eqs. (5) and (6). Maximizing Eq. (8) over f yields the optimal voluntary disclosure fine f, f = 1 + (1 q)(p H p L )F (1 q)(p H p L )c a + c c. (9) 2ty 10 In principle, a much lower administrative cost also arises after a voluntary disclosure or when assessing the tax returns of non-evaders. This cost is here normalized to zero. 14

15 The optimal fine for voluntary disclosures is higher when the fine for tax evasion (F ) is higher, and lower when the administrative costs associated with tax evasion or the compliance costs associated with preparing the voluntary disclosure are higher. When the difference between the detection probabilities in the two states of the world is large, the voluntary disclosure fine is higher, because the difference in detection probabilities increases the incentive for individuals to come clean. 11 For further interpretation, consider the cutoffs αh vd and α t as a function of the underlying parameters: αh vd 1 = ty(1 pf ) 2 (1 q)(p H p L )c a cc, (10) α t = ty(1 pf ) q(p H p L )c a 1 q 2 1 q cc. (11) Clearly, when there are no administrative costs or compliance costs, αh vd = α t (= α 0 ), i.e. the revenue-maximizing government then sets the fine so high that voluntary disclosure is not attractive for any evader. Comparing (10) and (11) shows that for voluntary disclosure to take place in equilibrium, it has to hold that c c c a (1 q)(p h p L ), as assumed in (2). Next, consider how the existence of a voluntary disclosure mechanism affects the tax evasion decision. Comparing α t with α 0 from the benchmark model without voluntary disclosure yields the following proposition: Proposition 1 The introduction of a voluntary disclosure program with a fine set optimally in the presence of administrative costs increases the number of individuals who evade taxes. Proof. Proof by contradiction: Assume α t < α 0. Then, from Eqs. (3) and (11), 0 > (p H p L )c a cc 1 q, which is a contradiction whenever parameters are such that voluntary disclosure takes place in equilibrium (i.e. when condition (2), which followed from comparing αh vd and α t, holds). To understand this result, consider how voluntary disclosure affects the tax evasion decision. Voluntary disclosure can be interpreted as an option that an individual may ( ) 11 df d(p H p L ) = (1 q) F ca 2ty, which is positive given the restrictions placed on parameters in (2). 15

16 exercise when the detection probability proves to be high. Without this option (i.e. in an economy without voluntary disclosure), individuals come to a decision about evading taxes based on the expected probability of detection, p. In contrast, if voluntary disclosure is possible, individuals anticipate that they can voluntarily disclose when the detection probability is high and thus decide about tax evasion based on the low detection probability p L and the voluntary disclosure fine f. In the extreme case of no voluntary disclosure fine (f = 1) and no compliance cost (c c = 0), they evade as if the detection probability was p L for sure. As more people evade taxes when the detection probability is lower, the possibility of voluntary disclosure increases the number of people who evade taxes to start with. The higher tax evasion when a voluntary disclosure program exists seems to be an argument against introducing such a program for a revenue-maximizing government. However, voluntary disclosure also has several other effects on tax revenues. With voluntary disclosure, more individuals pay tax (and the low fine f) in the state of the world with the high detection probability, and administrative costs are lower. In contrast, in the low detection probability state, is has clear negative effects as there is more evasion but no voluntary disclosures take place. To see the overall effect on tax revenues, I compare the equilibrium tax revenues T (derived by inserting the optimal fine f and αh vd and α t in Eq. 8) with Eq. (4), assuming a uniform distribution of α i also in this case. This shows T > T 0 [c c c a (p h p L )(1 q)] 2 q 4(1 q) > 0. (12) When administrative costs are positive, (12) is always fulfilled. Then, the existence of voluntary disclosure increases tax revenues net of administrative costs. Individuals disclosure of tax evasion leaves rents on the table, as the individuals have a lower cost of preparing the information for assessing previously evaded taxes than the tax authorities themselves. A voluntary disclosure program offers the government a way to share into these rents. Thus, the government decides to use voluntary disclosure despite the increase in tax evasion it causes. When there are no administrative costs (c a = 0), T = T In this case, the government optimally sets the voluntary disclosure fine so high that no voluntary dis- 12 When c a = 0, it follows from (2) that c c = 0 as it cannot be more expensive for tax evaders to gather information on their tax evasion than for the tax authorities. 16

17 closure takes place in equilibrium (see Eqs. 10 and 11). This behavior is optimal as an attractive voluntary disclosure program (i.e. a program with a fine sufficiently low that there is some uptake) would increase tax evasion, without the corresponding benefit of lower administrative costs. The following proposition summarizes these results: Proposition 2 If and only if there are administrative costs when assessing evaded taxes, the existence of a voluntary disclosure program raises expected tax revenues net of administrative costs. Proof. See Eq. (12). Intuitively, as long as administrative costs are positive, the voluntary disclosure mechanism generates efficiency gains in terms of reduced collection costs. The government can increase these efficiency gains by drawing people into the voluntary disclosure scheme, but this implies setting a low fine and foregoing additional tax revenue. In addition, a low fine implies that evasion becomes more attractive ex ante. 2.4 Administrative Costs Proposition 2 has shown that the government should implement a voluntary disclosure program if such a program is associated with lower administrative costs. If this is not the case, the government should not use this instrument (in the model, the government sets the fine such that no individual voluntarily discloses). The crucial assumption is that the administrative costs for the tax authorities are significantly lower when they assess evaded taxes on the basis of a voluntary disclosure, compared to a situation in which the tax authorities have detected the tax evasion themselves. As mentioned above, a voluntary disclosure usually has to contain all information necessary to assess taxes, and usually includes a revised tax return. Having this information prepared in a structured form likely lowers administrative costs for the tax authorities. However, it is also possible that most work arises when checking the accuracy of the disclosure. In that case, voluntary disclosures would not save administrative costs. Only the tax authorities themselves can answer which of these arguments dominates in reality. Therefore, I have carried out a survey among all regional tax offices in 17

18 Germany. I received answers from 12 of the 16 federal states. In eight cases, an agency at the level of the state answered, usually based on its own survey among the competent local tax authorities. From four other states, I directly received answers from competent local tax authorities, so that I have a total of 18 individual answers. My first question asked whether a voluntary disclosure increased or decreased the work time necessary to assess taxes, compared to a situation where the evasion has already been detected (e.g. by receiving whistle-blower information). A strong majority ( 60%) noted that administrative effort is significantly lower after a voluntary disclosure. When asked to quantify by how much the working time decreased, the answers ranged from 30% to 90%, with two thirds citing a decrease in the necessary work time above 80%. 13 Overall, a majority of respondents confirmed the argument made above that a voluntary disclosure strongly decreases the administrative effort necessary to assess previously evaded taxes. A second question gauged the absolute volume of administrative costs, by asking about the hours of work necessary to assess taxes after a voluntary disclosure. The answers ranged from minutes to several months, with an average of five days. This range has to be expected, given that the voluntary disclosures concern very different cases: They range from individuals who forgot to file a foreign account in a single year, to others who have multiple offshore accounts with holdings in several funds, where all realizations of capital gains in the last ten years have to be retraced. Interestingly, those competent local tax authorities who gave relatively high work time estimates also were more likely to answer that voluntary disclosure substantially lowers administrative costs. Possibly these local authorities deal predominantly with large cases, where the potential gains from voluntary disclosures are higher than for small cases. All in all, the survey evidence confirms that for many competent tax authorities, voluntary disclosures implies administrative cost that are substantially (up to 90%) lower than when assessing taxes based on whistle-blower information. In this situation the model predicts that government allows voluntary disclosures and fines them at a relatively low rate, as is the case in Germany (see Section 3.2). 13 About 16% of participants answered that the necessary work time increased, giving fact-checking efforts as the main reason. However, none of these respondents gave a numerical estimate by how much the required work time increased. The remainder stated that the work time does not change or did not answer, noting that the answer varies too much case by case. 18

19 3 Empirical Analysis In the following, I empirically test some aspects of voluntary disclosure. The main test considers Proposition 1, i.e. whether voluntary disclosure increases tax evasion. In a second test, I analyze a shock to the detection probability to gauge the size of the tax revenues effects. First, I study how the introduction of a voluntary disclosure program in the U.S. in 2009 affected tax evasion. In the model, this corresponds to comparing the case with voluntary disclosure with the one without it (see Figure 3). In Proposition 1, the model predicted that introducing voluntary disclosure increases tax evasion. To test this effect, I use data on offshore account balances, comparing the offshore deposits of U.S. residents with those from various control countries using a synthetic control method. Moreover, the model predicts that voluntary disclosures take place only when the high detection probability occurs, and that the voluntary disclosures lead to increased net tax revenues. An event that came close to such an exogenous increase in the detection probability (from the taxpayers point of view) is the acquisition of whistle-blower data of Swiss bank accounts by German tax authorities in early Figure 1 on page 6 confirmed that this acquisition was indeed associated with a strong increase in the use of voluntary disclosures. Section 3.2 uses this development to estimate the additional revenue raised by voluntary disclosures. This estimation is no true test of Proposition 2, as the counterfactual revenue without the voluntary disclosure program is unknown. In contrast, it compares the situation in which nature has drawn p L with one in which nature has drawn p H. It thus gives an estimate of the average size of voluntary disclosures, a helpful number when discussing the policy consequences of this analysis. 3.1 United States: Voluntary Disclosure and Tax Evasion Background The U.S. introduced a voluntary disclosure program in The IRS already experimented with voluntary disclosure programs in the first half of the twentieth century. However, since 1952, no formal policy regarding the civil penalties for intentional 19

20 tax evaders existed until In the criminal prosecution of tax evaders, individuals who came forward voluntarily have long been treated more favorably. Nevertheless, the 2009 initiative was the first large program to introduce a significantly more favorable civil tax penalty. 15 Therefore, the introduction of the voluntary disclosure program significantly affected the expected penalties perceived by U.S. residents. The 2009 program ran from mid-march till mid-october 2009 and was considered a success: About 15,000 taxpayers voluntarily disclosed prior tax evasion (U.S. Government Accountability Office, 2014). In February 2011, the IRS announced a follow-up program (the 2011 Offshore Voluntary Disclosure Initiative), which ended in mid-september Again, a large number of taxpayers (about 18,000) took advantage of this program. Ultimately, the IRS began an open-ended offshore voluntary disclosure program (OVDP) in January Table 2 provides an overview of some of the locations of foreign accounts declared in the 2009 program, showing that they referred to many different countries. [Table 2 about here.] All three initiatives had relatively similar requirements. 16 They referred specifically to unreported income from undisclosed offshore accounts for years after Individual taxpayers disclosing income in the program have to pay the full amount of tax, plus interest, and a monetary penalty of up to 25% of unpaid taxes. Moreover, there is an additional penalty of 20% (2009 program), 25% (2011 initiative) or 27.5% (2012 program) of the value of the assets in the foreign bank accounts. These penalties are significantly lower than the general punishments for tax evasion or failure to 14 An exception was a three-month program in 2003 (the 2003 offshore voluntary compliance initiative), which was aimed mostly at taxpayers who used offshore payment cards. As only 1,321 taxpayers used the 2003 initiative, I will in the following focus on the program started in For more information on the history of voluntary disclosure in the U.S. see Madison (2001) and U.S. Government Accountability Office (2013). 15 According to an example by the IRS, tax and penalty payments for a foreign account with $ 1 million and 5% yearly interest would accrue to $ 368,000 after a voluntary disclosure in the 2009 program, compared to $ 2,306,000 without the voluntary disclosure (IRS, 2009). 16 For details, see the IRS homepage at Program, and Offshore-Voluntary-Disclosure-Program. 20

21 declare foreign accounts. 17 Table 3 gives an overview of the taxes and penalties paid by participants in the 2009 OVDP. [Table 3 about here.] I use the introduction of the first program in 2009 to estimate how the existence of voluntary disclosure has affected tax evasion activities. While the 2009 program was only temporary, it marked a definite change in the IRS policy towards tax evaders: for the first time since 1952, a broad and encompassing scheme for repentant tax evaders was put in place. While no follow-up program was originally announced, the IRS formal acknowledgement that it treats tax evaders who come forward voluntarily in a more lenient way likely constitutes a significant shift in the perceived treatment of tax evaders. 18 It is therefore suited to test Proposition 1 despite formally being a temporary program Data and Descriptives By its nature, data on tax evasion is scarce. I therefore proxy for tax evasion using the deposits of U.S. residents in an aggregate of offshore banking centres. The Bank for International Settlements (BIS) provides this data on a quarterly basis. It collects the information necessary to compile this dataset from local banks in cooperation with the respective countries central banks. The offshore banking centers in this aggregate are 17 Civil penalties for tax evasion are the greater of $ 100,000 or 50% of the total balance of the foreign account. In addition, criminal penalties of up to $ 500,000 or up to 10 years of imprisonment are possible for the failure to file a report of foreign bank and financial accounts (OECD, 2010a). 18 In Question & Answer section for the 2009 voluntary disclosure program, the IRS states that taxpayers run a substantial risk that the uniform penalty structure described in the internal guidance will not be available past the 6-month deadline or that the terms will be less beneficial to taxpayers in the future (IRS, 2009). By noting that terms will be less beneficial, the IRS implicitly indicated the possibility that there would still be a somewhat beneficial treatment of voluntary disclosures after the formal program ended. At the very least, by officially offering more lenient penalties for offshore tax evaders for the first time in many decades, persons that were close to indifference whether to evade taxes likely concluded that other such possibilities might arise in the future. 19 The only country that recently introduced voluntary disclosure into the general law was Switzerland in However, as a general shift in the policy towards tax evaders took place in Switzerland around that time, it is not possible to analyze the consequences that the introduction of voluntary disclosure had there. 21

22 the Bahamas, Bahrain, Bermuda, Cayman Islands, Curaçao (from Q4 2010), Guernsey, Hong Kong, Isle of Man, Jersey, Macao, Netherlands Antilles (to Q3 2010), Panama, and Singapore. 20 The BIS designates these countries as offshore financial centers. At the same time, the economic literature identifies all of these countries as tax havens (Hines and Rice, 1994). In the following analysis, I make the assumption that the combination of bank secrecy and low tax rates makes these offshore financial centers a potentially attractive location for tax evaders. Table 4 gives an overview of the amount of assets held in these offshore banking centers, and shows that these assets make up a significant fraction of the overall assets U.S. residents hold abroad. [Table 4 about here.] Most deposits in offshore banking centers are not in the local currency. For each offshore banking center, the BIS provides data on the amount of deposits belonging to foreigners in either all currencies or foreign currencies. 21 Comparing these numbers shows that during the sample period, on average 94% of deposits belonging to foreigners in offshore banking centers were also in foreign currencies. 22 There are some potential issues with measuring tax evasion indirectly by foreign assets: First, it is not clear if these deposits really belong to individuals. Johannesen and Zucman (2014) show that households hold at least 50% of the tax haven deposits. Second, it is possible that individuals do pay tax on this income. There is, however, little reasons except tax evasion for individuals to hold assets in the very small countries in the offshore banking aggregate Unfortunately, data is only available for the aggregate of all offshore banking centers, not for individual offshore countries. 21 As this data is not specific to a counterparty location, it is available for each offshore banking center individually. 22 There is, however, substantial variation across the offshore banking centers. In the Carribean countries and Bahrain, almost all deposits are in foreign currencies. In Macao and Hong Kong, at least 85% of deposits are in foreign currencies. The British Crown dependencies Guernsey, Jersey and Isle of Man have substantially lower shares of foreign deposits in foreign currencies of 79% (Guernsey), 70% (Jersey) and 36% (Isle of Man). I discuss the results when matching also on exchange rates at the end of Section Johannesen and Zucman (2014) also show that tax treaties signed by a tax haven significantly decrease deposits held in this haven, confirming that tax haven deposits are a reasonable proxy for evaded taxes. I will discuss later how the signing of tax treaties may impact the results. 22